PhotoGPT Comprehensive Dreambooth and Text2Img Guide

This guide walks through the complete lifecycle of building a personalized model inside PhotoGPT and then extending it with the Text2Img studio. Whether you're creating professional headshots or thematic campaigns, these steps keep results consistent across every render.

1. Prepare a Focused Dataset

DreamBooth thrives on curated training data. Aim for 20–30 photos that:

• Highlight the subject's face and hairstyle with a mix of angles.
• Maintain consistent lighting (studio or natural) to reduce noise.
• Avoid occlusions such as sunglasses or busy backgrounds.

Upload images in the Training section and label the subject with a unique identifier (for example, brandx executive). PhotoGPT automatically balances your dataset and warns about duplicates.

2. Configure Training Settings

PhotoGPT exposes sensible defaults, but you can fine-tune:

• Steps – More steps improve fidelity but increase training time. Start with 800–1,000.
• Learning rate – Leave at the suggested value unless you see overfitting.
• Class images – Enable if you want PhotoGPT to blend in reference poses during training.

Kick off training and monitor progress from the dashboard. We'll notify you when the model is ready to use.

3. Generate with Text2Img

Once your DreamBooth session completes, open the Text2Img tab and select your custom model. Every prompt should include the identifier you defined during training, e.g. brandx executive in a modern office portrait.

Start with these prompt building blocks:

brandx executive, cinematic lighting, 85mm portrait photography

Layer additional descriptors to match campaign goals—wardrobe, setting, or emotion. Because PhotoGPT stores the full DreamBooth checkpoint, generations remain true to the subject while still responding to creative direction.

4. Iterate with Variations

Use the "Variations" control to spin up alternate crops, backgrounds, or expressions without rewriting prompts. When you find a winner, mark it as a favorite so teammates can reuse it later.

5. Export & Share

Download finals in full resolution or trigger instant resizing for social formats. The export panel also lets you generate comparison grids that showcase before/after transformations—a perfect deliverable for clients.

Best Practices

• Refresh your model with new uploads every few months to keep styling current.
• Store successful prompts in the workspace library so your team can build on them.
• Combine DreamBooth portraits with preset backgrounds for consistent brand collateral.

By pairing DreamBooth training with the text-to-image studio, PhotoGPT becomes a turnkey system for branded imagery.